| Summary: | 碩士 === 義守大學 === 資訊工程學系碩士班 === 98 === Nowadays, there are several popular fire detection devices, including smoke sensor, heat sensor and infrared sensor. However, each of these detection devices has its own limitations, such as installed location, the distance between detection device and fire, and fire level. For example, heat sensor works when the surrounding temperature reaches to some certain level. Therefore, an efficient fire alarm system is urgently expected to detect the unwanted presence of fire. This thesis proposes a fast fire alarm system based on image processing method to detect the fire. We first segment a large number of fire regions from video sequences to collect fire-pixel samples. Then a learning-based approach is established so that we can determine weather each pixel is a fire pixel by comparing its color with the pre-trained database. An image pattern is classified as a fire region if the ratio of the number of fire-pixels of two frames is larger than a certain threshold. Finally, we compute the fire growth rate between two frames to predict the firing condition, and start the fire alarm if the fire incident is identified. The experimental result of indicates that the proposed method achieves efficient fire detection and can be wildly used for various environments.
|
|---|
